Isabelle Guyon's home page
Isabelle Guyon
ClopiNet
955 Creston road
Berkeley, CA 94708, USA
1+ (510) 524-6211
isabelle@clopinet.com
My Projects:
Pattern Classification and Machine Learning:
- Optimum Margin Classifiers and Support Vector
Machines: A new technique of classification, which leaves the largest
possible margin on either side of the decision boundary, therefore reducing
classification errors. Our invention was granted a US patent.
Take a look at the impressive SVM application
list. For further information, consult the kernel machines web site, the
support vector machines
web site, and the list of available
SVM software.
- Optimum test set size: What is the
minimum number of samples needed to estimate the performance of your classifier
with a certain accuracy? Lambert Schomaker wrote an Applet
based on the results of our research.
- Variable and feature selection:
Methods for selecting just a few variables relevant for classification,
with applications to gene selection, drug discovery, text mining, and
others. See the special issue of JMLR
we co-edited and the benchmark on feature
extraction we organized. See also the book
we edited and the class we taught.
Biometrics, Genomics, Proteomics and Cancer research:
- Fingerprint verification: Your fingerprint may be stored on your credit card in the
future. We devised a method to represent fingerprints in a compact way using
directional and frequency maps.
- Writer identification: Handwriting
has always been used by forensic experts for law enforcement purpose. Fighting
against terrorism has recently incrased the needs for accurate writer indentification.
We contributed to the design of the Wanda XML
format, a new standard to store and annotate handwriting pieces of evidence.
- Gene selection: A project to discover
genes that may be connected to cancer or other diseases by analyzing patterns
of gene expression obtained from DNA microarrays. A patent is pending
on our invention of RFE-SVM (Recursive Feature Elimination SVM).
- Diagnosis from protein profiles:
Gene activity monitors only indirectly protein regulation. Disease states
can be better assessed by profiling protein amounts directly. We use antibody
array and mass-spectrometry data to help diagnosing disease, including cancer.
Handwriting Recognition and Pen Computing:
- UNIPEN: A project of data
exchange and benchmark for on-line handwritten data.
- On-line Handwriting Recognizer:
A program that can recognize handprinted and cursive words using Time
Delay Neural Networks (TDNN) and Hidden Markov Models (HMM). A US patent
was granted to our application of TDNNs.
- Cursive handwriting teacher: A program
which measures how well formed your cursive handwriting is.
- Handwriting Synthesizer: A program
that synthesizes text with your own handwriting, given an ASCII file.
- Language model: A program which
learns the statistics of English using Variable Memory Length Markov Models
(VLMM) and Weighted Finite State Transductions (WFST). Such models are
used, in particular, as handwriting recognition postprocessors.
- Wanda: A framework for
writer identification with forensic applications.
Organization of workshops and challenges
CHALLENGES IN MACHINE LEARNING
- NIPS 92: learning theory workshop on capacity
(with Esther Levin and Michael Kearns).
- NIPS
01: variable and feature selection workshop (with David Lewis).
- NIPS
02: negative results and open problems workshop.
- NIPS
03: feature extraction workshop (with Richard Caruana, Kristin Bennett,
and Masoud Nikravesh) and feature selection challenge
(with Steve Gunn, Asa Ben Hur, Andre Elisseeff, and Gideon Dror).
- WCCI06:
model selection special session and performance prediction challenge
(with Amir Reza saffari Azar, Gideon Dror, Olivier Guyon, and many others).
- NIPS 06:
multilevel inference workshop and model selection game (with Gavin Cawley,
Amir Reza saffari Azar, Gideon Dror, Olivier Guyon, and many others).
- NIPS 06: causality and
feature selection (with Andre Elisseeff and Constantin Aliferis)
- IJCNN 07:
data representation discovery workshop and agnostic learning vs. prior knowledge
challenge (with Gavin Cawley, Amir Reza saffari Azar, Gideon Dror, Olivier
Guyon, and many others).
- Causal discovery workbench
(with Constantin Aliferis, André Elisseeff, Greg Cooper, Peter Spirtes,
and many others).
Teaching
Some of my friends:
Former advisors and research directors
Former students
Colleagues and collaborators
- Prof. Dr. Lambert Schomaker,
Univ. Groningen, The Netherlands.
- Katrin Franke,
Fraunhofer IPK, Berlin, Germany.
- Dr. André
Elisseeff, IBM Zürich, Switzerland.
- Dr. Masoud
Nikravesh, UC Berkeley, USA.
- Dr. René
Doursat, University of Nevada, Reno, USA.
- Dr. Steve Gunn,
Univ. Southamptom, UK.
- Prof. Joachim
Buhmann, ETH, Zurich.
- Dr. Gideon Dror, Academic
College of Tel-Aviv-Yaffo, Israel.
- Amir Reza Saffari Azar Alamdary,
Graz University of Technology, Austria.
- Prof. Gavin Cawley, University
of East Anglia, UK.
- Prof.
Constantin Aliferis, Vanderbilt University, Nashville, USA.
My tools:
My family:
My family comes from France, except my husband Bernhard who comes
from Switzerland.
Here are the two of us the
day of our wedding in Zuerich.
I live in Berkeley, California, with my husband and three children. For more family pictures,
we have a BIG album.
The rest of my family lives in Paris, France.
I am often on the phone with my mother Denise. She studied architecture and
became a real estate agent. I hear more rarely about my father Jean who does not like calling or writing.
He is an architect. On the picture, he is shown in his country house near
Paris, with his second wife Anita, my niece Solene and my grand-mother Madeleine
Lizer, who was a painter.
My sister Alice is a biologist
and a poet. On the picture she is shown with my grand-father
on my mother side: Maurice Passy.
She has now a large family. My
brother Olivier is web designer and
Rock-and-Roll guitar player.